Teaching vision-language models semiotics: toward a socio-semiotic framework for multimodal AI | Synapse